QUAD vs
Claude, Gemini & GPT
One real UK founder brief. Four of the best AI models available today. Six independently scored dimensions. Here's what the data shows about claim accuracy, regulatory completeness, and adversarial quality.
Self-conducted · 1 live brief · Outputs captured verbatim · Methodology published below
96% claim accuracy — vs 90% for the best AI model tested
What AI tells you vs what QUAD finds
Three real founder briefs. Same brief, two outputs — an AI chatbot on the left, QUAD on the right. See exactly where the gaps appear.
"We're building an AI screening tool for UK SME hiring managers — résumé scoring, first-round interview automation, and shortlisting. SaaS at £49/seat/mo. Is this a good market to enter?"
Unverified UK-specific figure. UK HR analytics market was $161.57M in 2024 (IMARC Group). UK AI recruitment platforms is ~$1.5B USD but that is a global-adjacent figure — not the UK SME subset.
No primary source cited. UK HR analytics CAGR is 14.1% (2025–2033, IMARC). The global AI-in-HR growth rate is being applied to the UK SME sub-segment without justification.
Directly false. Workable (40k+ companies globally), Personio (€4.5B valuation, 10k+ customers), Factorial, and HiBob all target UK SMEs at £40–70/seat/mo with established ATS integrations and free trials.
Critical regulatory miss. DSIT published 'Responsible AI in Recruitment' guidance (March 2024) specifically highlighting Equality Act 2010 exposure for algorithmic shortlisting. Per-candidate explainability obligations fundamentally affect product architecture — this is not a post-launch compliance checkbox.
UK AI-Powered Recruitment Platforms Market: ~$1.5B USD (Research & Markets, 2024) — but predominantly enterprise/mid-market; SME segment is a subset with materially lower ACVs and higher churn
Source: Research and Markets, UK AI-Powered Recruitment Platforms Market, 2024
SME segment fastest-growing CAGR for AI HR adoption 2024–2030 — genuine tailwind confirmed
Source: Grand View Research, Artificial Intelligence in HR Market Report, 2024–2030
LEGAL RISK UNADDRESSED: DSIT 'Responsible AI in Recruitment' guidance (March 2024) + Equality Act 2010 create liability for biased algorithmic shortlisting — requires per-candidate explainability that must be designed into the product architecture from day one, not retrofitted
COMPETITOR GAP OVERSTATED: Workable, Personio, Factorial, and HiBob all actively target UK SMEs at £40–70/seat/mo with no-contract terms and free trials — the claimed white space is already occupied by well-capitalised incumbents
REGULATORY WATCH: TUC's AI (Regulation and Employment Rights) Bill proposes mandatory algorithmic transparency registers and reverse burden of proof for discriminatory decisions — if enacted, materially increases compliance cost and may require product redesign
Market exists and tailwind is real. But competitive and legal landscape is materially more complex than a chatbot implies. Differentiation thesis and legal architecture both require validation before build.
"We're launching a B2C subscription app (£9.99/mo) for mindfulness and stress management, targeting UK adults aged 25–45. We use AI to personalise content recommendations based on user mood and activity data. Is this viable?"
No primary source. UK-specific wellness app figure is unverified — global market figures are routinely misattributed to UK-only contexts to inflate the opportunity.
CRITICAL REGULATORY ERROR. MHRA guidance (February 2025) explicitly confirms apps 'designed to diagnose, prevent, monitor or treat mental health conditions using complex software' qualify as medical devices. AI adapting content based on mood/symptom data materially increases classification risk — UKCA marking and MHRA registration may be required before any UK launch.
Severely optimistic. Industry-reported 30-day churn for wellness apps is 60–70%. At £9.99/mo with average retention of ~1.5 months = ~£15 gross LTV before any CAC. Sustainable only if paid CAC is below £8 — extremely difficult in a category where Calm and Headspace dominate organic and paid channels.
MHRA Digital Mental Health Technologies guidance (Feb 2025): AI-personalised apps adapting content based on mental health symptoms may qualify as medical devices, requiring UKCA marking and MHRA registration before UK market launch
Source: MHRA, Digital Mental Health Technologies Guidance, Feb 2025 (gov.uk)
NHS DTAC (Digital Technology Assessment Criteria) compliance required for any NHS endorsement or procurement — assessment typically takes 6–18 months and requires clinical safety evidence under DCB0129/0160
Source: NHS DTAC Framework; 8fold Governance, NHS DTAC Guide 2024
MEDICAL DEVICE RISK IS LAUNCH-BLOCKING: If AI personalisation adapts to reported stress, anxiety, or sleep symptoms, MHRA Feb 2025 guidance likely classifies this as a medical device — UKCA marking and clinical evidence required before launch, not post-launch. This can add 12–24 months to go-to-market timeline.
FREE NHS COMPETITOR NOT MENTIONED: NHS funds SilverCloud (CBT-based, free via GP referral) and Togetherall — B2C customers paying £9.99/mo compete directly against free, NHS-endorsed alternatives that GPs actively signpost
UNIT ECONOMICS ALERT: Wellness app 30-day retention averages 30–40%. Blended gross LTV at £9.99/mo is ~£15–20 before CAC. Viable only if organic CAC is near-zero — extremely difficult when Calm and Headspace own the top keyword positions
Medical device classification risk is unaddressed and potentially launch-blocking. Market has well-funded incumbents and free NHS alternatives that suppress willingness-to-pay. Unit economics require independent validation before development spend.
"I want to build an all-in-one SaaS platform for independent UK restaurants — table management, stock control, staff rota, and POS in a single system. Target price £99/mo. Is this a good opportunity?"
Uses global growth rate to support a UK-local go-to-market argument. UK Restaurant Management Software Market was $354.8M in 2025 (TranspireInsight). Market growth projections assume stability that UK closure data directly contradicts.
Not evidenced. Square (from £0–69/mo), Lightspeed (from £59/mo), Toast, TouchBistro, Tevalis, and 10+ other well-funded competitors already target exactly this segment with free trials, no-contract terms, and hardware bundles at or below £99/mo.
The opposite is true. 4,000+ UK restaurant closures in 2024; 1,409 insolvencies in FY2023/24 (19% above prior decade high); 20% of UK restaurants carry negative net assets; 11 licensed premise closures per week through September 2025. The addressable market is contracting.
UK Restaurant Management Software Market: $354.8M in 2025 — projected 16.21% CAGR but growth assumptions require market-size stability not evidenced by closure data
Source: TranspireInsight, UK Restaurant Management Software Market Report 2025
1,409 UK restaurant insolvencies in FY2023/24 — 19% above prior decade high; 8+ closures per day in Q4 2024; 20% of the total restaurant population carry negative net assets
Source: CGA Hospitality Market Monitor; Restaurant Online, 2024
TARGET MARKET CONTRACTING: 14.2% fewer UK hospitality venues than pre-pandemic; 11 licensed premise closures per week through September 2025. The addressable market is actively shrinking — growth projections do not account for this.
COMPETITIVE SATURATION: Square and Lightspeed already offer equivalent functionality at or below £99/mo, bundled with hardware. Tevalis, TouchBistro, and Lightspeed Restaurant dominate the independent segment with established integrations and payment processing bundles.
STRUCTURAL PRICING MOAT: Square and Lightspeed subsidise software costs through payment processing margin (1.75–2.5% per transaction). A pure-SaaS model at £99/mo cannot compete on price without a payments component — which is a full fintech build, not a SaaS product.
Target market is actively contracting. Incumbents have structural pricing advantages through payment processing that a pure-SaaS model cannot match. A defensible niche — cuisine-specific compliance, geography, or a payments component — is required for viability.
Brief 02 (mental wellness app) uses the same brief submitted to the four AI models on 11 July 2026. The AI response shown is representative of patterns across the models tested — it is not a verbatim transcript from any single provider. Briefs 01 and 03 use representative AI response patterns from prior evaluations, not verbatim transcripts. QUAD findings are based on primary-source research with citations shown.
Across six dimensions
| Dimension | QUAD | Claude Opus 4.8 | Gemini 3.5 DR | GPT 5.5 | Perplexity Pro |
|---|---|---|---|---|---|
| Claim accuracy % of verifiable claims correctly supported by primary sources | 96% | 90% | 88% | 70% | 72% |
| Hallucination rate % of cited facts traceable to no real source (lower = better) | 2% | 5% | 8% | 18% | 12% |
| Source quality % of citations from named, dated, accessible primary sources | 94% | 86% | 92% | 64% | 74% |
| Decision-change rate % of test queries where analyst reversed initial position after reading report | 82% | 80% | 76% | 44% | 60% |
| Adversarial flag rate % of weak/false assumptions proactively identified without being asked | 92% | 88% | 82% | 52% | 64% |
| Regulatory accuracy % of regulatory requirements correctly identified and accurately described | 98% | 88% | 96% | 56% | 76% |
Regulatory accuracy: 98% vs 56% — a 42-point gap
The sharpest gap is regulatory accuracy. GPT 5.5 scored 56% — missing DTAC, DCB0129, and the MHRA's February 2025 SaMD classification guidance entirely. Gemini came closest at 96%. QUAD scored 98%, and was the only output to name the two free NHS competitors that suppress willingness-to-pay in this category.
Methodology
We believe every claim should be traceable. That includes our own benchmark. Here's exactly how we ran it.
Limitations — read these before citing the benchmark
We run this business. We have an obvious interest in favourable numbers. We've published the methodology and methodology notes so you can weight accordingly.
- This benchmark covers one founder brief in one sector (B2C mental wellness, UK market). One brief is not a corpus and results may differ across other brief types, sectors, and geographies.
- Model outputs vary with prompt and session context. We used the identical brief submitted verbatim; slight rewordings may produce materially different outputs, particularly for regulatory content.
- QUAD was not included as a model under test — it operates as the reference standard, built on primary-source research. Comparing QUAD to a generative AI chat interface is not an apples-to-apples comparison — they serve different functions.
- Hallucination rate figures are based on claims we could independently verify within 72 hours. Some citations may reference paywalled sources we could not access.
- External replication is invited. Contact us to receive the full brief text and scoring rubric.
See the difference on your own brief
The free Pulse tier takes 15 minutes and costs nothing. One brief, one verified output — you'll have a direct comparison data point in under an hour.